Research Commonsense Knowledge Acquisition Approaches Based On Sentence With Serial Verbs In Chinese

Commonsense knowledge acquisition is not only an important research but also a long-standing challengein artificial intelligence.It aims to build huge commonsense knowledge bases for realistic applications and to implement intelligentsystems.Prerequisite commonsenseknowledgeand consequencecommonsense knowledge,as two different kinds of commonsense knowledge,have its important practical application value in question answering,natural language understanding,information retrieval and other fields.However,due to the implicity and fundamentality of commonsense knowledgeautomatic acquisition methods usually do not collect the implicit knowledge,which is missing in the raw text or is often ignored by knowledge engineers.Sentence with serial verbs is a common sentence pattern in Chinese which contains two or more predicate verbs.And the two verbs areinterdependent,they have goal,casual,manner and other semantic relations.A predicate verb is an event and sentence with serial verbs describes multi-events,so sentence with serial verbs contains a fund of knowledge.Attractively,the massivenesseasy accessibility and holistic domain independence of sentence with serial verbsmake it easyto acquire large-scale commonsense knowledge from it.Motivated by this backgroundwe systematically researched the theory and method ofcommonsense knowledgeacquisitionbased on sentence with serial verbs of Chinese.The main work of this paper is mainly reflected in the three following aspects:Firstly,extraction of sentence with serial verbs identification method.A method is proposed which combines rule methods and statistical methods to identify sentence with serial verbs.Rulesare constructed based on formal features and semantic roles of sentence with serial verbs,and statistical methods are proposed to calculate the characteristics of the intermediate words between two predicate verbs to achieve automatic identification of sentence with serial verbs.The experiment result shows that the method based on rules and statistic has accuracy rate of 75.48%,compared with identification method based on rules base to improve 23.7%.Secondly,extraction ofgeneration method of serial-verb grammar.Ruleswith semantic features and syntactic structure of sentence with serial verbsare proposed based on the event semantic.Serial-verb grammarprovides the theory foundation for commonsenseknowledgeacquisition.Lastly,extraction ofcommonsense knowledgeacquisition based on sentence with serial verbs.Four kinds methods of commonsense knowledge acquisition are proposed,which are based on SVS semantic grammar,serial verbs,the angles of knowledge and the types of sentence with serial verbs.Then seven kinds of interaction questions are designed to suggest and lead knowledge engineer to acquire commonsense knowledge.In order to exam the acceptability and efficiency of the interaction process,we proposed the acceptability and validity based on the binomial distribution hypothesis to evaluate the questions in the quantitative form.Experimental results show that the method arrives at accuracy rate of 92.5%.